Intrinsic Images and their Applications in Intelligent Systems

Abstract

The overall goal of the thesis is to research intelligent systems and to provide one more innovative piece in the puzzle towards general artificial intelligence. Because one quickly realizes the importance of computer vision for this endeavor, and in there specifically the need to understand the 3D world through their 2D projections into images, we thoroughly investigate the field of intrinsic images in this thesis and improve the intrinsic decomposition of arbitrary images to enable smarter intelligent systems. We demonstrate the utilization of such a decomposition in the task of relighting, where the intrinsic structure is shown to improve results.